Ruby

=begin
Sorry, ruby -v: ruby 1.9.2dev (2010-06-01 revision 28116) [x86_64-darwin10.3.0]

=end

=begin
Sorry again, this should be in ruby-1.9... :/
=end

=begin
The "problem" is that Object#<=> calls #== internally:

rb_obj_cmp(VALUE obj1, VALUE obj2)
{
if (obj1 == obj2 || rb_equal(obj1, obj2)) // <- Calls rb_equal
return INT2FIX(0);
return Qnil;
}

VALUE
rb_equal(VALUE obj1, VALUE obj2)
{
VALUE result;

 if (obj1 == obj2) return Qtrue; result = rb_funcall(obj1, id_eq, 1, obj2); // <- Calls #== if (RTEST(result)) return Qtrue; return Qfalse; 

}
=end

=begin
Ok, but then imho the Exception is inappropriate and most people don't expect a SystemStackError when they found a method implementation missing. For most people it isn't obvious (see commit message in the ref link) why including Comparable and not implementing <=> (because they want to delegate the call through the method_missing hook) should cause a SystemStackError.

It should respect the method_missing hook. But thanks for the answer (Magnus).
=end

=begin
Hi,

In message "Re: [ruby-core:30727] [Bug #3420] Module#method calling <=> causes SystemStackError"
on Fri, 11 Jun 2010 08:30:26 +0900, Florian Aßmann redmine@ruby-lang.org writes:

|Ok, but then imho the Exception is inappropriate and most people don't expect a SystemStackError when they found a method implementation missing. For most people it isn't obvious

Could you elaborate your opinion bit more? Most people don't redefine
comparison operator, I expect.

matz. 

=end

=begin
Please let me ask, why as from 1.9.2:

Sat Oct 24 13:38:45 2009 Yukihiro Matsumoto

* object.c (rb_obj_cmp): defines Object#<=>. [ruby-core:24063] 

Before that all ruby implementations did:

ruby-1.8.7-p249 >
BrokenComparable = Module.new { def ==(other) self <=> other end }
BrokenClass = Class.new { include BrokenComparable }
BrokenClass.new == nil
NoMethodError: undefined method <=>' for #<BrokenClass:0x10043ef08> from (irb):1:in =='
from (irb):3

Which is a lot more verbose on what happened and allowed people to hook into method missing and delegate it to a wrapped object as a SystemStackError. Now, if there is a global defined <=> for all classes which mix in the kernel module (all but BasicObject) I'd suspect it should either just work (not calling rb_equal from rb_obj_cmp) as long people are used to 1.8.x/1.9.1 or be more verbose on what happened.

Removing the call of rb_equal in rb_obj_cmp would not just save the extra test of obj identity but also work for all objects; which seems to me like the intended behavior when adding a method like <=> to almost all objects.
Otherwise it should be stated in the docs of Comparable what happens if one does not implement the <=> method and what the currently undocumented Object#<=> actually does since a SystemStackError only adds confusion and isn't obvious at all.

Regards
Florian
=end